X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flov%2Flov_obd.c;h=3186999eed994e94c60bd0fd09746fe61c84a40e;hb=87c86d444e61e38d6454bba5700ba966dc1ac83d;hp=2a0b76d3bb6f861cccbcbe0b07b87576d36b1971;hpb=08e9a34c98c7b1ad67744a18ae57ba0b40cf4a97;p=fs%2Flustre-release.git diff --git a/lustre/lov/lov_obd.c b/lustre/lov/lov_obd.c index 2a0b76d..3186999 100644 --- a/lustre/lov/lov_obd.c +++ b/lustre/lov/lov_obd.c @@ -1,210 +1,317 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * lov/lov.c - * - * Copyright (C) 2002 Cluster File Systems, Inc. - * Author: Phil Schwan + * Copyright (C) 2002, 2003 Cluster File Systems, Inc. + * Author: Phil Schwan * Peter Braam + * Mike Shaver + * + * This file is part of Lustre, http://www.lustre.org. + * + * Lustre is free software; you can redistribute it and/or + * modify it under the terms of version 2 of the GNU General Public + * License as published by the Free Software Foundation. + * + * Lustre is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. * - * This code is issued under the GNU General Public License. - * See the file COPYING in this distribution + * You should have received a copy of the GNU General Public License + * along with Lustre; if not, write to the Free Software + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ -#define EXPORT_SYMTAB +#ifndef EXPORT_SYMTAB +# define EXPORT_SYMTAB +#endif #define DEBUG_SUBSYSTEM S_LOV - +#ifdef __KERNEL__ #include #include +#include +#include +#include +#include +#include +#else +#include +#endif + #include #include #include #include +#include #include #include #include -#include -#include +#include +#include + +#include "lov_internal.h" /* obd methods */ -static int lov_connect(struct lustre_handle *conn, struct obd_device *obd, - obd_uuid_t cluuid, struct recovd_obd *recovd, - ptlrpc_recovery_cb_t recover) +#define MAX_STRING_SIZE 128 +static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt, + int activate, struct obd_connect_data *conn_data, + unsigned long connect_flags) { - struct ptlrpc_request *req = NULL; + struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" }; + struct obd_uuid *tgt_uuid = &tgt->uuid; + +#ifdef __KERNEL__ + struct proc_dir_entry *lov_proc_dir; +#endif struct lov_obd *lov = &obd->u.lov; - struct client_obd *mdc = &lov->mdcobd->u.cli; - struct lov_desc *desc = &lov->desc; - struct lustre_handle mdc_conn; - obd_uuid_t *uuidarray; - int rc, rc2, i; + struct lustre_handle conn = {0, }; + struct obd_device *tgt_obd; + int rc; + ENTRY; - MOD_INC_USE_COUNT; - rc = class_connect(conn, obd, cluuid); - if (rc) { - MOD_DEC_USE_COUNT; - RETURN(rc); + tgt_obd = class_find_client_obd(tgt_uuid, LUSTRE_OSC_NAME, + &obd->obd_uuid); + + if (!tgt_obd) { + CERROR("Target %s not attached\n", tgt_uuid->uuid); + RETURN(-EINVAL); } - /* We don't want to actually do the underlying connections more than - * once, so keep track. */ - lov->refcount++; - if (lov->refcount > 1) + if (!tgt_obd->obd_set_up) { + CERROR("Target %s not set up\n", tgt_uuid->uuid); + RETURN(-EINVAL); + } + + if (activate) { + tgt_obd->obd_no_recov = 0; + ptlrpc_activate_import(tgt_obd->u.cli.cl_import); + } + + if (tgt_obd->u.cli.cl_import->imp_invalid) { + CERROR("not connecting OSC %s; administratively " + "disabled\n", tgt_uuid->uuid); + rc = obd_register_observer(tgt_obd, obd); + if (rc) { + CERROR("Target %s register_observer error %d; " + "will not be able to reactivate\n", + tgt_uuid->uuid, rc); + } RETURN(0); + } - /* retrieve LOV metadata from MDS */ - rc = obd_connect(&mdc_conn, lov->mdcobd, NULL, recovd, recover); + rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, conn_data, + connect_flags); if (rc) { - CERROR("cannot connect to mdc: rc = %d\n", rc); - GOTO(out_conn, rc); + CERROR("Target %s connect error %d\n", tgt_uuid->uuid, rc); + RETURN(rc); } + tgt->ltd_exp = class_conn2export(&conn); - rc = mdc_getlovinfo(obd, &mdc_conn, &req); - rc2 = obd_disconnect(&mdc_conn); + rc = obd_register_observer(tgt_obd, obd); if (rc) { - CERROR("cannot get lov info %d\n", rc); - GOTO(out_conn, rc); + CERROR("Target %s register_observer error %d\n", + tgt_uuid->uuid, rc); + obd_disconnect(tgt->ltd_exp, 0); + tgt->ltd_exp = NULL; + RETURN(rc); } - if (rc2) { - CERROR("error disconnecting from MDS %d\n", rc2); - GOTO(out_conn, rc = rc2); + tgt->active = 1; + lov->desc.ld_active_tgt_count++; + +#ifdef __KERNEL__ + lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); + if (lov_proc_dir) { + struct obd_device *osc_obd = class_conn2obd(&conn); + struct proc_dir_entry *osc_symlink; + char name[MAX_STRING_SIZE + 1]; + + LASSERT(osc_obd != NULL); + LASSERT(osc_obd->obd_type != NULL); + LASSERT(osc_obd->obd_type->typ_name != NULL); + name[MAX_STRING_SIZE] = '\0'; + snprintf(name, MAX_STRING_SIZE, "../../../%s/%s", + osc_obd->obd_type->typ_name, + osc_obd->obd_name); + osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir, + name); + if (osc_symlink == NULL) { + CERROR("could not register LOV target " + "/proc/fs/lustre/%s/%s/target_obds/%s\n", + obd->obd_type->typ_name, obd->obd_name, + osc_obd->obd_name); + lprocfs_remove(lov_proc_dir); + lov_proc_dir = NULL; + } } +#endif - /* sanity... */ - if (req->rq_repmsg->bufcount < 2 || - req->rq_repmsg->buflens[0] < sizeof(*desc)) { - CERROR("LOV desc: invalid descriptor returned\n"); - GOTO(out_conn, rc = -EINVAL); - } + RETURN(0); +} - memcpy(desc, lustre_msg_buf(req->rq_repmsg, 0), sizeof(*desc)); - lov_unpackdesc(desc); +static int lov_connect(struct lustre_handle *conn, struct obd_device *obd, + struct obd_uuid *cluuid, struct obd_connect_data *data, + unsigned long flags) +{ +#ifdef __KERNEL__ + struct proc_dir_entry *lov_proc_dir; +#endif + struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; + struct obd_export *exp; + int rc, rc2, i; + ENTRY; - if (req->rq_repmsg->buflens[1] < sizeof(*uuidarray)*desc->ld_tgt_count){ - CERROR("LOV desc: invalid uuid array returned\n"); - GOTO(out_conn, rc = -EINVAL); - } + rc = class_connect(conn, obd, cluuid); + if (rc) + RETURN(rc); - mdc->cl_max_mds_easize = lov_mds_md_size(desc->ld_tgt_count); - mdc->cl_max_ost_easize = lov_stripe_md_size(desc->ld_tgt_count); + exp = class_conn2export(conn); - if (memcmp(obd->obd_uuid, desc->ld_uuid, sizeof(desc->ld_uuid))) { - CERROR("LOV desc: uuid %s not on mds device (%s)\n", - obd->obd_uuid, desc->ld_uuid); - GOTO(out_conn, rc = -EINVAL); + /* We don't want to actually do the underlying connections more than + * once, so keep track. */ + lov->refcount++; + if (lov->refcount > 1) { + class_export_put(exp); + RETURN(0); } - if (desc->ld_tgt_count > 1000) { - CERROR("LOV desc: target count > 1000 (%d)\n", - desc->ld_tgt_count); - GOTO(out_conn, rc = -EINVAL); +#ifdef __KERNEL__ + lov_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry, + NULL, NULL); + if (IS_ERR(lov_proc_dir)) { + CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.", + obd->obd_type->typ_name, obd->obd_name); + lov_proc_dir = NULL; } +#endif - /* Because of 64-bit divide/mod operations only work with a 32-bit - * divisor in a 32-bit kernel, we cannot support a stripe width - * of 4GB or larger on 32-bit CPUs. - */ - if ((desc->ld_default_stripe_count ? - desc->ld_default_stripe_count : desc->ld_tgt_count) * - desc->ld_default_stripe_size > ~0UL) { - CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n", - desc->ld_default_stripe_size, - desc->ld_default_stripe_count ? - desc->ld_default_stripe_count : desc->ld_tgt_count,~0UL); - GOTO(out_conn, rc = -EINVAL); + /* connect_flags is the MDS number, save for use in lov_add_obd */ + lov->lov_connect_flags = flags; + for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) { + if (obd_uuid_empty(&tgt->uuid)) + continue; + rc = lov_connect_obd(obd, tgt, 0, data, flags); + if (rc) + GOTO(out_disc, rc); } - lov->bufsize = sizeof(struct lov_tgt_desc) * desc->ld_tgt_count; - OBD_ALLOC(lov->tgts, lov->bufsize); - if (!lov->tgts) { - CERROR("Out of memory\n"); - GOTO(out_conn, rc = -ENOMEM); + class_export_put(exp); + RETURN (0); + + out_disc: +#ifdef __KERNEL__ + if (lov_proc_dir) + lprocfs_remove(lov_proc_dir); +#endif + + while (i-- > 0) { + struct obd_uuid uuid; + --tgt; + --lov->desc.ld_active_tgt_count; + tgt->active = 0; + /* save for CERROR below; (we know it's terminated) */ + uuid = tgt->uuid; + rc2 = obd_disconnect(tgt->ltd_exp, 0); + if (rc2) + CERROR("error: LOV target %s disconnect on OST idx %d: " + "rc = %d\n", uuid.uuid, i, rc2); } + class_disconnect(exp, 0); + RETURN (rc); +} - uuidarray = lustre_msg_buf(req->rq_repmsg, 1); - for (i = 0; i < desc->ld_tgt_count; i++) - memcpy(lov->tgts[i].uuid, uuidarray[i], sizeof(*uuidarray)); +static int lov_disconnect_obd(struct obd_device *obd, + struct lov_tgt_desc *tgt, + unsigned long flags) +{ +#ifdef __KERNEL__ + struct proc_dir_entry *lov_proc_dir; +#endif + struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp); + struct lov_obd *lov = &obd->u.lov; + int rc; + ENTRY; - for (i = 0; i < desc->ld_tgt_count; i++) { - struct obd_device *tgt = class_uuid2obd(uuidarray[i]); - if (!tgt) { - CERROR("Target %s not attached\n", uuidarray[i]); - GOTO(out_disc, rc = -EINVAL); - } - if (!(tgt->obd_flags & OBD_SET_UP)) { - CERROR("Target %s not set up\n", uuidarray[i]); - GOTO(out_disc, rc = -EINVAL); - } - rc = obd_connect(&lov->tgts[i].conn, tgt, NULL, recovd, - recover); - if (rc) { - CERROR("Target %s connect error %d\n", - uuidarray[i], rc); - GOTO(out_disc, rc); +#ifdef __KERNEL__ + lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); + if (lov_proc_dir) { + struct proc_dir_entry *osc_symlink; + + osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name); + if (osc_symlink) { + lprocfs_remove(osc_symlink); + } else { + CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n", + obd->obd_type->typ_name, obd->obd_name, + osc_obd->obd_name); } - desc->ld_active_tgt_count++; - lov->tgts[i].active = 1; + } +#endif + if (obd->obd_no_recov) { + /* Pass it on to our clients. + * XXX This should be an argument to disconnect, + * XXX not a back-door flag on the OBD. Ah well. + */ + if (osc_obd) + osc_obd->obd_no_recov = 1; } - out: - ptlrpc_req_finished(req); - return rc; + obd_register_observer(tgt->ltd_exp->exp_obd, NULL); + rc = obd_disconnect(tgt->ltd_exp, flags); + if (rc) { + if (tgt->active) { + CERROR("Target %s disconnect error %d\n", + tgt->uuid.uuid, rc); + } + rc = 0; + } - out_disc: - while (i-- > 0) { - desc->ld_active_tgt_count--; - lov->tgts[i].active = 0; - rc2 = obd_disconnect(&lov->tgts[i].conn); - if (rc2) - CERROR("LOV Target %s disconnect error: rc = %d\n", - uuidarray[i], rc2); + if (tgt->active) { + tgt->active = 0; + lov->desc.ld_active_tgt_count--; } - OBD_FREE(lov->tgts, lov->bufsize); - out_conn: - class_disconnect(conn); - goto out; + tgt->ltd_exp = NULL; + RETURN(0); } -static int lov_disconnect(struct lustre_handle *conn) +static int lov_disconnect(struct obd_export *exp, unsigned long flags) { - struct obd_device *obd = class_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); +#ifdef __KERNEL__ + struct proc_dir_entry *lov_proc_dir; +#endif struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; int rc, i; + ENTRY; if (!lov->tgts) goto out_local; - /* Only disconnect the underlying laters on the final disconnect. */ + /* Only disconnect the underlying layers on the final disconnect. */ lov->refcount--; if (lov->refcount != 0) goto out_local; - for (i = 0; i < lov->desc.ld_tgt_count; i++) { - if (!lov->tgts[i].active) { - CERROR("Skipping disconnect for inactive OSC %s\n", - lov->tgts[i].uuid); - continue; - } - - lov->desc.ld_active_tgt_count--; - lov->tgts[i].active = 0; - rc = obd_disconnect(&lov->tgts[i].conn); - if (rc) { - CERROR("Target %s disconnect error %d\n", - lov->tgts[i].uuid, rc); - RETURN(rc); - } + for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) { + if (tgt->ltd_exp) + lov_disconnect_obd(obd, tgt, flags); } - OBD_FREE(lov->tgts, lov->bufsize); - lov->bufsize = 0; - lov->tgts = NULL; +#ifdef __KERNEL__ + lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); + if (lov_proc_dir) { + lprocfs_remove(lov_proc_dir); + } else { + CERROR("/proc/fs/lustre/%s/%s/target_obds missing.", + obd->obd_type->typ_name, obd->obd_name); + } +#endif + out_local: - rc = class_disconnect(conn); - if (!rc) - MOD_DEC_USE_COUNT; - return rc; + rc = class_disconnect(exp, 0); + RETURN(rc); } /* Error codes: @@ -212,48 +319,43 @@ static int lov_disconnect(struct lustre_handle *conn) * -EINVAL : UUID can't be found in the LOV's target list * -ENOTCONN: The UUID is found, but the target connection is bad (!) * -EBADF : The UUID is found, but the OBD is the wrong type (!) - * -EALREADY: The OSC is already marked (in)active */ -static int lov_set_osc_active(struct lov_obd *lov, obd_uuid_t uuid, +static int lov_set_osc_active(struct lov_obd *lov, struct obd_uuid *uuid, int activate) { - struct obd_device *obd; + struct lov_tgt_desc *tgt; int i, rc = 0; ENTRY; CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n", - lov, uuid, activate); + lov, uuid->uuid, activate); spin_lock(&lov->lov_lock); - for (i = 0; i < lov->desc.ld_tgt_count; i++) - if (strncmp(uuid, lov->tgts[i].uuid, - sizeof(lov->tgts[i].uuid)) == 0) + for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) { + if (tgt->ltd_exp == NULL) + continue; + + CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n", + i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie); + + if (obd_uuid_equals(uuid, &tgt->uuid)) break; + } if (i == lov->desc.ld_tgt_count) GOTO(out, rc = -EINVAL); - obd = class_conn2obd(&lov->tgts[i].conn); - if (obd == NULL) { - LBUG(); - GOTO(out, rc = -ENOTCONN); - } - - CDEBUG(D_INFO, "Found OBD %p type %s\n", obd, obd->obd_type->typ_name); - if (strcmp(obd->obd_type->typ_name, "osc") != 0) { - LBUG(); - GOTO(out, rc = -EBADF); - } - if (lov->tgts[i].active == activate) { - CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd, - activate ? "" : "in"); - GOTO(out, rc = -EALREADY); + if (tgt->active == activate) { + CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid, + activate ? "" : "in"); + GOTO(out, rc); } - CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd, activate ? "" : "in"); + CDEBUG(D_INFO, "Marking OSC %s %sactive\n", uuid->uuid, + activate ? "" : "in"); - lov->tgts[i].active = activate; + tgt->active = activate; if (activate) lov->desc.ld_active_tgt_count++; else @@ -265,850 +367,1455 @@ static int lov_set_osc_active(struct lov_obd *lov, obd_uuid_t uuid, return rc; } -static int lov_setup(struct obd_device *obd, obd_count len, void *buf) +static int lov_notify(struct obd_device *obd, struct obd_device *watched, + int active, void *data) { - struct obd_ioctl_data* data = buf; - struct lov_obd *lov = &obd->u.lov; - int rc = 0; + struct obd_uuid *uuid; + int rc; ENTRY; - if (data->ioc_inllen1 < 1) { - CERROR("osc setup requires an MDC UUID\n"); - RETURN(-EINVAL); + if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) { + CERROR("unexpected notification of %s %s!\n", + watched->obd_type->typ_name, + watched->obd_name); + return -EINVAL; } + uuid = &watched->u.cli.cl_import->imp_target_uuid; - if (data->ioc_inllen1 > 37) { - CERROR("mdc UUID must be 36 characters or less\n"); - RETURN(-EINVAL); + /* Set OSC as active before notifying the observer, so the + * observer can use the OSC normally. + */ + rc = lov_set_osc_active(&obd->u.lov, uuid, active); + if (rc) { + CERROR("%sactivation of %s failed: %d\n", + active ? "" : "de", uuid->uuid, rc); + RETURN(rc); } - spin_lock_init(&lov->lov_lock); - lov->mdcobd = class_uuid2obd(data->ioc_inlbuf1); - if (!lov->mdcobd) { - CERROR("LOV %s cannot locate MDC %s\n", obd->obd_uuid, - data->ioc_inlbuf1); - rc = -EINVAL; - } + if (obd->obd_observer) + /* Pass the notification up the chain. */ + rc = obd_notify(obd->obd_observer, watched, active, data); + RETURN(rc); } - -/* the LOV expects oa->o_id to be set to the LOV object id */ -static int lov_create(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md **ea) +int lov_attach(struct obd_device *dev, obd_count len, void *data) { - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_stripe_md *lsm; - struct lov_oinfo *loi; - struct obdo *tmp; - int ost_count, ost_idx, i, rc = 0; - ENTRY; - - LASSERT(ea); - - if (!export) - RETURN(-EINVAL); - - tmp = obdo_alloc(); - if (!tmp) - RETURN(-ENOMEM); + struct lprocfs_static_vars lvars; + int rc; - lov = &export->exp_obd->u.lov; + lprocfs_init_vars(lov, &lvars); + rc = lprocfs_obd_attach(dev, lvars.obd_vars); + if (rc == 0) { +#ifdef __KERNEL__ + struct proc_dir_entry *entry; - spin_lock(&lov->lov_lock); - ost_count = lov->desc.ld_tgt_count; - oa->o_easize = lov_stripe_md_size(ost_count); - - lsm = *ea; - if (!lsm) { - OBD_ALLOC(lsm, oa->o_easize); - if (!lsm) { - spin_unlock(&lov->lov_lock); - GOTO(out_tmp, rc = -ENOMEM); + entry = create_proc_entry("target_obd_status", 0444, + dev->obd_proc_entry); + if (entry == NULL) { + rc = -ENOMEM; + } else { + entry->proc_fops = &lov_proc_target_fops; + entry->data = dev; } +#endif } + return rc; +} + +int lov_detach(struct obd_device *dev) +{ + return lprocfs_obd_detach(dev); +} - LASSERT(oa->o_valid & OBD_MD_FLID); - lsm->lsm_magic = LOV_MAGIC; - lsm->lsm_mds_easize = lov_mds_md_size(ost_count); - lsm->lsm_object_id = oa->o_id; - if (!lsm->lsm_stripe_count) - lsm->lsm_stripe_count = lov->desc.ld_default_stripe_count; - if (!lsm->lsm_stripe_count) - lsm->lsm_stripe_count = lov->desc.ld_active_tgt_count; - else if (lsm->lsm_stripe_count > lov->desc.ld_active_tgt_count) - lsm->lsm_stripe_count = lov->desc.ld_active_tgt_count; +static int lov_setup(struct obd_device *obd, obd_count len, void *buf) +{ + struct lov_obd *lov = &obd->u.lov; + struct lustre_cfg *lcfg = buf; + struct lov_desc *desc; + int count; + ENTRY; - if (!lsm->lsm_stripe_size) - lsm->lsm_stripe_size = lov->desc.ld_default_stripe_size; + if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { + CERROR("LOV setup requires a descriptor\n"); + RETURN(-EINVAL); + } + desc = (struct lov_desc *)lustre_cfg_string(lcfg, 1); + if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) { + CERROR("descriptor size wrong: %d > %d\n", + (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1)); + RETURN(-EINVAL); + } + /* Because of 64-bit divide/mod operations only work with a 32-bit * divisor in a 32-bit kernel, we cannot support a stripe width * of 4GB or larger on 32-bit CPUs. */ - if (lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL) { + + count = desc->ld_default_stripe_count; + if (count && (count * desc->ld_default_stripe_size) > ~0UL) { CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n", - lsm->lsm_stripe_size, lsm->lsm_stripe_count, ~0UL); - spin_unlock(&lov->lov_lock); - GOTO(out_free, rc = -EINVAL); - } - - lsm->lsm_ost_count = ost_count; - if (!lsm->lsm_stripe_offset) { - int mult = lsm->lsm_object_id * lsm->lsm_stripe_count; - int stripe_offset = mult % ost_count; - int sub_offset = (mult / ost_count) % lsm->lsm_stripe_count; - - lsm->lsm_stripe_offset = stripe_offset + sub_offset; + desc->ld_default_stripe_size, count, ~0UL); + RETURN(-EINVAL); } - - /* Pick the OSTs before we release the lock */ - ost_idx = lsm->lsm_stripe_offset; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - do { - ost_idx = (ost_idx + 1) % ost_count; - } while (!lov->tgts[ost_idx].active); - CDEBUG(D_INFO, "Using ost_idx %d (uuid %s)\n", ost_idx, - lov->tgts[ost_idx].uuid); - loi->loi_ost_idx = ost_idx; + if (desc->ld_tgt_count > 0) { + lov->bufsize= sizeof(struct lov_tgt_desc) * desc->ld_tgt_count; + } else { + lov->bufsize = sizeof(struct lov_tgt_desc) * LOV_MAX_TGT_COUNT; } - - spin_unlock(&lov->lov->lock); - - CDEBUG(D_INODE, "allocating %d subobjs for objid "LPX64" at idx %d\n", - lsm->lsm_stripe_count,lsm->lsm_object_id,lsm->lsm_stripe_offset); - - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - struct lov_stripe_md obj_md; - struct lov_stripe_md *obj_mdp = &obj_md; - - ost_idx = loi->loi_ost_idx; - - /* create data objects with "parent" OA */ - memcpy(tmp, oa, sizeof(*tmp)); - tmp->o_easize = sizeof(struct lov_stripe_md); - rc = obd_create(&lov->tgts[ost_idx].conn, tmp, &obj_mdp); - if (rc) { - CERROR("error creating objid "LPX64" sub-object on " - "OST idx %d: rc = %d\n", oa->o_id, ost_idx, rc); - GOTO(out_cleanup, rc); - } - loi->loi_id = tmp->o_id; - loi->loi_size = tmp->o_size; - CDEBUG(D_INODE, "objid "LPX64" has subobj "LPX64" at idx %d\n", - lsm->lsm_object_id, loi->loi_id, ost_idx); + OBD_ALLOC(lov->tgts, lov->bufsize); + if (lov->tgts == NULL) { + lov->bufsize = 0; + CERROR("couldn't allocate %d bytes for target table.\n", + lov->bufsize); + RETURN(-EINVAL); } - *ea = lsm; + desc->ld_tgt_count = 0; + desc->ld_active_tgt_count = 0; + lov->desc = *desc; + spin_lock_init(&lov->lov_lock); + sema_init(&lov->lov_llog_sem, 1); - out_tmp: - obdo_free(tmp); - return rc; + RETURN(0); +} - out_cleanup: - while (i-- > 0) { - int err; +static int lov_cleanup(struct obd_device *obd, int flags) +{ + struct lov_obd *lov = &obd->u.lov; - --loi; - /* destroy already created objects here */ - memcpy(tmp, oa, sizeof(*tmp)); - tmp->o_id = loi->loi_id; - err = obd_destroy(&lov->tgts[loi->loi_ost_idx].conn, tmp, NULL); - if (err) - CERROR("Failed to uncreate objid "LPX64" subobj " - LPX64" on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, - err); - } - out_free: - OBD_FREE(lsm, oa->o_easize); - goto out_tmp; + OBD_FREE(lov->tgts, lov->bufsize); + RETURN(0); } -static int lov_destroy(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +static int +lov_add_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_oinfo *loi; - int rc = 0, i; + struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; + int rc; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea for destruction\n"); + CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d\n", + uuidp->uuid, index, gen); + + if ((index < 0) || (index >= LOV_MAX_TGT_COUNT)) { + CERROR("request to add OBD %s at invalid index: %d\n", + uuidp->uuid, index); RETURN(-EINVAL); } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); + if (gen <= 0) { + CERROR("request to add OBD %s with invalid generation: %d\n", + uuidp->uuid, gen); RETURN(-EINVAL); } - if (!export || !export->exp_obd) - RETURN(-ENODEV); - - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; - rc = obd_destroy(&lov->tgts[loi->loi_ost_idx].conn, &tmp, NULL); - if (rc) - CERROR("Error destroying objid "LPX64" subobj "LPX64 - " on OST idx %d\n: rc = %d", - oa->o_id, loi->loi_id, loi->loi_ost_idx, rc); + tgt = lov->tgts + index; + if (!obd_uuid_empty(&tgt->uuid)) { + CERROR("OBD already assigned at LOV target index %d\n", + index); + RETURN(-EEXIST); } - RETURN(rc); -} -/* compute object size given "stripeno" and the ost size */ -static obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size, - int stripeno) -{ - unsigned long ssize = lsm->lsm_stripe_size; - unsigned long swidth = ssize * lsm->lsm_stripe_count; - unsigned long stripe_size; - obd_size lov_size; + tgt->uuid = *uuidp; + /* XXX - add a sanity check on the generation number. */ + tgt->ltd_gen = gen; - if (ost_size == 0) - return 0; + if (index >= lov->desc.ld_tgt_count) + lov->desc.ld_tgt_count = index + 1; - /* do_div(a, b) returns a % b, and a = a / b */ - stripe_size = do_div(ost_size, ssize); + CDEBUG(D_CONFIG, "idx: %d ltd_gen: %d ld_tgt_count: %d\n", + index, tgt->ltd_gen, lov->desc.ld_tgt_count); - if (stripe_size) - lov_size = ost_size * swidth + stripeno * ssize + stripe_size; - else - lov_size = (ost_size - 1) * swidth + (stripeno + 1) * ssize; + if (lov->refcount == 0) + RETURN(0); - return lov_size; -} + if (tgt->ltd_exp) { + struct obd_device *osc_obd; -static void lov_merge_attrs(struct obdo *tgt, struct obdo *src, obd_flag valid, - struct lov_stripe_md *lsm, int stripeno, int *new) -{ - if (*new) { - obdo_cpy_md(tgt, src, valid); - if (valid & OBD_MD_FLSIZE) - tgt->o_size = lov_stripe_size(lsm,src->o_size,stripeno); - *new = 0; - } else { - if (valid & OBD_MD_FLSIZE) { - /* this handles sparse files properly */ - obd_size lov_size; + osc_obd = class_exp2obd(tgt->ltd_exp); + if (osc_obd) + osc_obd->obd_no_recov = 0; + } - lov_size = lov_stripe_size(lsm, src->o_size, stripeno); - if (lov_size > tgt->o_size) - tgt->o_size = lov_size; - } - if (valid & OBD_MD_FLBLOCKS) - tgt->o_blocks += src->o_blocks; - if (valid & OBD_MD_FLCTIME && tgt->o_ctime < src->o_ctime) - tgt->o_ctime = src->o_ctime; - if (valid & OBD_MD_FLMTIME && tgt->o_mtime < src->o_mtime) - tgt->o_mtime = src->o_mtime; + rc = lov_connect_obd(obd, tgt, 1, NULL, lov->lov_connect_flags); + if (rc) + GOTO(out, rc); + + if (obd->obd_observer) { + /* tell the mds_lov about the new target */ + rc = obd_notify(obd->obd_observer, tgt->ltd_exp->exp_obd, 1, + (void *)index); } + + GOTO(out, rc); + out: + if (rc && tgt->ltd_exp != NULL) + lov_disconnect_obd(obd, tgt, 0); + return rc; } -static int lov_getattr(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +static int +lov_del_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_oinfo *loi; - int rc = 0, i; - int new = 1; + struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; + int count = lov->desc.ld_tgt_count; + int rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); - } + CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d\n", + uuidp->uuid, index, gen); - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); + if (index >= count) { + CERROR("LOV target index %d >= number of LOV OBDs %d.\n", + index, count); RETURN(-EINVAL); } - if (!export || !export->exp_obd) - RETURN(-ENODEV); + tgt = lov->tgts + index; - lov = &export->exp_obd->u.lov; - oa->o_size = 0; - oa->o_blocks = 0; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - int err; + if (obd_uuid_empty(&tgt->uuid)) { + CERROR("LOV target at index %d is not setup.\n", index); + RETURN(-EINVAL); + } - if (loi->loi_id == 0) - continue; + if (!obd_uuid_equals(uuidp, &tgt->uuid)) { + CERROR("LOV target UUID %s at index %d doesn't match %s.\n", + tgt->uuid.uuid, index, uuidp->uuid); + RETURN(-EINVAL); + } - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; + if (tgt->ltd_exp) { + struct obd_device *osc_obd; - err = obd_getattr(&lov->tgts[loi->loi_ost_idx].conn, &tmp,NULL); - if (err) { - CERROR("Error getattr objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); - if (!rc) - rc = err; - continue; /* XXX or break? */ + osc_obd = class_exp2obd(tgt->ltd_exp); + if (osc_obd) { + osc_obd->obd_no_recov = 1; + rc = obd_llog_finish(osc_obd, &osc_obd->obd_llogs, 1); + if (rc) + CERROR("osc_llog_finish error: %d\n", rc); } - lov_merge_attrs(oa, &tmp, tmp.o_valid, lsm, i, &new); + lov_disconnect_obd(obd, tgt, 0); } + + /* XXX - right now there is a dependency on ld_tgt_count being the + * maximum tgt index for computing the mds_max_easize. So we can't + * shrink it. */ + + /* lt_gen = 0 will mean it will not match the gen of any valid loi */ + memset(tgt, 0, sizeof(*tgt)); + + CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n", + tgt->uuid.uuid, index, tgt->ltd_gen, tgt->ltd_exp, tgt->active); + RETURN(rc); } -static int lov_setattr(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +static int lov_process_config(struct obd_device *obd, obd_count len, void *buf) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_oinfo *loi; - int rc = 0, i; + struct lustre_cfg *lcfg = buf; + struct obd_uuid obd_uuid; + int cmd; + int index; + int gen; + int rc = 0; ENTRY; - /* Note that this code is currently unused, hence LBUG(), just - * to know when/if it is ever revived that it needs cleanups. - */ - LBUG(); - - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); + switch(cmd = lcfg->lcfg_command) { + case LCFG_LOV_ADD_OBD: + case LCFG_LOV_DEL_OBD: { + if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) + GOTO(out, rc = -EINVAL); + + obd_str2uuid(&obd_uuid, lustre_cfg_string(lcfg, 1)); + + if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1) + GOTO(out, rc = -EINVAL); + if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1) + GOTO(out, rc = -EINVAL); + if (cmd == LCFG_LOV_ADD_OBD) + rc = lov_add_obd(obd, &obd_uuid, index, gen); + else + rc = lov_del_obd(obd, &obd_uuid, index, gen); + GOTO(out, rc); } + default: { + CERROR("Unknown command: %d\n", lcfg->lcfg_command); + GOTO(out, rc = -EINVAL); - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); } + } +out: + RETURN(rc); +} - if (!export || !export->exp_obd) - RETURN(-ENODEV); +#ifndef log2 +#define log2(n) ffz(~(n)) +#endif + +static int lov_clear_orphans(struct obd_export *export, + struct obdo *src_oa, + struct lov_stripe_md **ea, + struct obd_trans_info *oti) +{ + struct lov_obd *lov; + struct obdo *tmp_oa; + struct obd_uuid *ost_uuid = NULL; + int rc = 0, i; + ENTRY; - /* size changes should go through punch and not setattr */ - LASSERT(!(oa->o_valid & OBD_MD_FLSIZE)); + LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS && + src_oa->o_flags == OBD_FL_DELORPHAN); lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { + + tmp_oa = obdo_alloc(); + if (tmp_oa == NULL) + RETURN(-ENOMEM); + + if (src_oa->o_valid & OBD_MD_FLINLINE) { + ost_uuid = (struct obd_uuid *)src_oa->o_inline; + CDEBUG(D_HA, "clearing orphans only for %s\n", + ost_uuid->uuid); + } + + for (i = 0; i < lov->desc.ld_tgt_count; i++) { int err; + struct lov_stripe_md obj_md; + struct lov_stripe_md *obj_mdp = &obj_md; + + /* + * if called for a specific target, we don't care if it is not + * active. + */ + if (lov->tgts[i].active == 0 && ost_uuid == NULL) { + CDEBUG(D_HA, "lov idx %d inactive\n", i); + continue; + } - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; + if (ost_uuid && !obd_uuid_equals(ost_uuid, &lov->tgts[i].uuid)) + continue; - err = obd_setattr(&lov->tgts[loi->loi_ost_idx].conn, &tmp,NULL); + /* + * setting up objid OSS objects should be destroyed starting + * from it. + */ + memcpy(tmp_oa, src_oa, sizeof(*tmp_oa)); + tmp_oa->o_valid |= OBD_MD_FLID; + tmp_oa->o_id = oti->oti_objid[i]; + + /* XXX: LOV STACKING: use real "obj_mdp" sub-data */ + err = obd_create(lov->tgts[i].ltd_exp, tmp_oa, NULL, 0, + &obj_mdp, oti); if (err) { - CERROR("Error setattr objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); - if (!rc) - rc = err; + /* + * this export will be disabled until it is recovered, + * and then orphan recovery will be completed. + */ + CERROR("error in orphan recovery on OST idx %d/%d: " + "rc = %d\n", i, lov->desc.ld_tgt_count, err); } + + if (ost_uuid) + break; } + obdo_free(tmp_oa); RETURN(rc); } -static int lov_open(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +/* the LOV expects oa->o_id to be set to the LOV object id */ +static int +lov_create(struct obd_export *exp, struct obdo *src_oa, + void *acl, int acl_size, struct lov_stripe_md **ea, + struct obd_trans_info *oti) { - struct obdo *tmp; - struct obd_export *export = class_conn2export(conn); + struct lov_request_set *set = NULL; + struct list_head *pos; struct lov_obd *lov; - struct lov_oinfo *loi; - int new = 1; - int rc = 0, i; + int rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea for opening\n"); + LASSERT(ea != NULL); + if (exp == NULL) RETURN(-EINVAL); + + if ((src_oa->o_valid & OBD_MD_FLFLAGS) && + src_oa->o_flags == OBD_FL_DELORPHAN) { + rc = lov_clear_orphans(exp, src_oa, ea, oti); + RETURN(rc); } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + lov = &exp->exp_obd->u.lov; + if (!lov->desc.ld_active_tgt_count) + RETURN(-EIO); + + LASSERT(oti->oti_flags & OBD_MODE_CROW); + + /* main creation loop */ + rc = lov_prep_create_set(exp, ea, src_oa, oti, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + struct lov_request *req = + list_entry(pos, struct lov_request, rq_link); + + /* XXX: LOV STACKING: use real "obj_mdp" sub-data */ + rc = obd_create(lov->tgts[req->rq_idx].ltd_exp, + req->rq_oa, NULL, 0, &req->rq_md, oti); + lov_update_create_set(set, req, rc); } + rc = lov_fini_create_set(set, ea); + RETURN(rc); +} + +#define lsm_bad_magic(LSMP) \ +({ \ + struct lov_stripe_md *_lsm__ = (LSMP); \ + int _ret__ = 0; \ + if (!_lsm__) { \ + CERROR("LOV requires striping ea\n"); \ + _ret__ = 1; \ + } else if (_lsm__->lsm_magic != LOV_MAGIC) { \ + CERROR("LOV striping magic bad %#x != %#x\n", \ + _lsm__->lsm_magic, LOV_MAGIC); \ + _ret__ = 1; \ + } \ + _ret__; \ +}) + +static int lov_destroy(struct obd_export *exp, struct obdo *oa, + struct lov_stripe_md *lsm, struct obd_trans_info *oti) +{ + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov; + int rc = 0; + ENTRY; + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); - if (!export || !export->exp_obd) + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - tmp = obdo_alloc(); - if (!tmp) - RETURN(-ENOMEM); + lov = &exp->exp_obd->u.lov; + rc = lov_prep_destroy_set(exp, oa, lsm, oti, &set); + if (rc) + RETURN(rc); - lov = &export->exp_obd->u.lov; - oa->o_size = 0; - oa->o_blocks = 0; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { + list_for_each (pos, &set->set_list) { int err; + req = list_entry(pos, struct lov_request, rq_link); - /* create data objects with "parent" OA */ - memcpy(tmp, oa, sizeof(*tmp)); - tmp->o_id = loi->loi_id; - - err = obd_open(&lov->tgts[loi->loi_ost_idx].conn, tmp, NULL); - if (err) { - CERROR("Error open objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, lsm->lsm_oinfo[i].loi_id, - loi->loi_ost_idx, rc); + /* XXX update the cookie position */ + oti->oti_logcookies = set->set_cookies + req->rq_stripe; + rc = obd_destroy(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, + NULL, oti); + err = lov_update_common_set(set, req, rc); + if (rc) { + CERROR("error: destroying objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + set->set_oa->o_id, req->rq_oa->o_id, + req->rq_idx, rc); if (!rc) rc = err; } - - lov_merge_attrs(oa, tmp, tmp->o_valid, lsm, i, &new); } - /* FIXME: returning an error, but having opened some objects is a bad - * idea, since they will likely never be closed. We either - * need to not return an error if _some_ objects could be - * opened, and leave it to read/write to return -EIO (with - * hopefully partial error status) or close all opened objects - * and return an error. I think the former is preferred. - */ - obdo_free(tmp); + lov_fini_destroy_set(set); RETURN(rc); } -static int lov_close(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +static int lov_getattr(struct obd_export *exp, struct obdo *oa, + struct lov_stripe_md *lsm) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; struct lov_obd *lov; - struct lov_oinfo *loi; - int rc = 0, i; + int err = 0, rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + + rc = lov_prep_getattr_set(exp, oa, lsm, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + req = list_entry(pos, struct lov_request, rq_link); + + CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " + "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id, + req->rq_idx); + + rc = obd_getattr(lov->tgts[req->rq_idx].ltd_exp, + req->rq_oa, NULL); + err = lov_update_common_set(set, req, rc); + if (err) { + CERROR("error: getattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + set->set_oa->o_id, req->rq_oa->o_id, + req->rq_idx, err); + break; + } + } + + rc = lov_fini_getattr_set(set); + if (err) + rc = err; + RETURN(rc); +} + +static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, void *data, + int rc) +{ + struct lov_request_set *lovset = (struct lov_request_set *)data; + ENTRY; + + /* don't do attribute merge if this aysnc op failed */ + if (rc) { + lovset->set_completes = 0; + lov_fini_getattr_set(lovset); + } else { + rc = lov_fini_getattr_set(lovset); } + RETURN (rc); +} - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); +static int lov_getattr_async(struct obd_export *exp, struct obdo *oa, + struct lov_stripe_md *lsm, + struct ptlrpc_request_set *rqset) +{ + struct lov_request_set *lovset; + struct lov_obd *lov; + struct list_head *pos; + struct lov_request *req; + int rc = 0; + ENTRY; + + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + + rc = lov_prep_getattr_set(exp, oa, lsm, &lovset); + if (rc) + RETURN(rc); + + CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n", + lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size); + + list_for_each (pos, &lovset->set_list) { + req = list_entry(pos, struct lov_request, rq_link); + + CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " + "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id, + req->rq_idx); + rc = obd_getattr_async(lov->tgts[req->rq_idx].ltd_exp, + req->rq_oa, NULL, rqset); + if (rc) { + CERROR("error: getattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + lovset->set_oa->o_id, req->rq_oa->o_id, + req->rq_idx, rc); + GOTO(out, rc); + } + lov_update_common_set(lovset, req, rc); } + + LASSERT(rc == 0); + LASSERT (rqset->set_interpret == NULL); + rqset->set_interpret = lov_getattr_interpret; + rqset->set_arg = (void *)lovset; + RETURN(rc); +out: + LASSERT(rc); + lov_fini_getattr_set(lovset); + RETURN(rc); +} - if (!export || !export->exp_obd) +static int lov_setattr(struct obd_export *exp, struct obdo *src_oa, + struct lov_stripe_md *lsm, struct obd_trans_info *oti) +{ + struct lov_request_set *set; + struct lov_obd *lov; + struct list_head *pos; + struct lov_request *req; + int err = 0, rc = 0; + ENTRY; + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - int err; + LASSERT(!(src_oa->o_valid & ~(OBD_MD_FLID|OBD_MD_FLTYPE | OBD_MD_FLMODE| + OBD_MD_FLATIME | OBD_MD_FLMTIME | + OBD_MD_FLCTIME | OBD_MD_FLFLAGS | + OBD_MD_FLSIZE | OBD_MD_FLGROUP | + OBD_MD_FLUID | OBD_MD_FLGID))); - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; + LASSERT(!(src_oa->o_valid & OBD_MD_FLGROUP) || src_oa->o_gr > 0); + + lov = &exp->exp_obd->u.lov; + rc = lov_prep_setattr_set(exp, src_oa, lsm, NULL, &set); + if (rc) + RETURN(rc); - err = obd_close(&lov->tgts[loi->loi_ost_idx].conn, &tmp, NULL); + list_for_each (pos, &set->set_list) { + req = list_entry(pos, struct lov_request, rq_link); + + rc = obd_setattr(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, + NULL, NULL); + err = lov_update_common_set(set, req, rc); if (err) { - CERROR("Error close objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); + CERROR("error: setattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + set->set_oa->o_id, req->rq_oa->o_id, + req->rq_idx, err); if (!rc) rc = err; } } + err = lov_fini_setattr_set(set); + if (!rc) + rc = err; RETURN(rc); } -#ifndef log2 -#define log2(n) ffz(~(n)) -#endif - -#warning FIXME: merge these two functions now that they are nearly the same - -/* compute ost offset in stripe "stripeno" corresponding to offset "lov_off" */ -static obd_off lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off, - int stripeno) +static int lov_revalidate_policy(struct lov_obd *lov, struct lov_stripe_md *lsm) { - unsigned long ssize = lsm->lsm_stripe_size; - unsigned long swidth = ssize * lsm->lsm_stripe_count; - unsigned long stripe_off, this_stripe; + static int next_idx = 0; + struct lov_tgt_desc *tgt; + int i, count; - if (lov_off == OBD_OBJECT_EOF || lov_off == 0) - return lov_off; + /* XXX - we should do something clever and take lsm + * into account but just do round robin for now. */ - /* do_div(a, b) returns a % b, and a = a / b */ - stripe_off = do_div(lov_off, swidth); + /* last_idx must always be less that count because + * ld_tgt_count currently cannot shrink. */ + count = lov->desc.ld_tgt_count; - this_stripe = stripeno * ssize; - if (stripe_off <= this_stripe) - stripe_off = 0; - else { - stripe_off -= this_stripe; - - if (stripe_off > ssize) - stripe_off = ssize; + for (i = next_idx, tgt = lov->tgts + i; i < count; i++, tgt++) { + if (tgt->active) { + next_idx = (i + 1) % count; + RETURN(i); + } } + for (i = 0, tgt = lov->tgts; i < next_idx; i++, tgt++) { + if (tgt->active) { + next_idx = (i + 1) % count; + RETURN(i); + } + } - return lov_off * ssize + stripe_off; + RETURN(-EIO); } -/* compute which stripe number "lov_off" will be written into */ -static int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off) +static int lov_revalidate_md(struct obd_export *exp, struct obdo *src_oa, + struct lov_stripe_md *ea, + struct obd_trans_info *oti) { - unsigned long ssize = lsm->lsm_stripe_size; - unsigned long swidth = ssize * lsm->lsm_stripe_count; - unsigned long stripe_off; + struct obd_export *osc_exp; + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_stripe_md *lsm = ea; + struct lov_stripe_md obj_md; + struct lov_stripe_md *obj_mdp = &obj_md; + struct lov_oinfo *loi; + struct obdo *tmp_oa; + int ost_idx, updates = 0, i; + ENTRY; + + tmp_oa = obdo_alloc(); + if (tmp_oa == NULL) + RETURN(-ENOMEM); - stripe_off = do_div(lov_off, swidth); + loi = lsm->lsm_oinfo; + for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) { + int rc; + if (!obd_uuid_empty(&lov->tgts[loi->loi_ost_idx].uuid)) + continue; - return stripe_off / ssize; -} + ost_idx = lov_revalidate_policy(lov, lsm); + if (ost_idx < 0) { + /* FIXME: punt for now. */ + CERROR("lov_revalidate_policy failed; no active " + "OSCs?\n"); + continue; + } + + /* create a new object */ + memcpy(tmp_oa, src_oa, sizeof(*tmp_oa)); + /* XXX: LOV STACKING: use real "obj_mdp" sub-data */ + osc_exp = lov->tgts[ost_idx].ltd_exp; + rc = obd_create(osc_exp, tmp_oa, NULL, 0, &obj_mdp, oti); + if (rc) { + CERROR("error creating new subobj at idx %d; " + "rc = %d\n", ost_idx, rc); + continue; + } + if (oti->oti_objid) + oti->oti_objid[ost_idx] = tmp_oa->o_id; + loi->loi_id = tmp_oa->o_id; + loi->loi_gr = tmp_oa->o_gr; + loi->loi_ost_idx = ost_idx; + loi->loi_ost_gen = lov->tgts[ost_idx].ltd_gen; + CDEBUG(D_INODE, "replacing objid "LPX64" subobj "LPX64 + " with idx %d gen %d.\n", lsm->lsm_object_id, + loi->loi_id, ost_idx, loi->loi_ost_gen); + updates = 1; + } + /* If we got an error revalidating an entry there's no need to + * cleanup up objects we allocated here because the bad entry + * still points to a deleted OST. */ + + obdo_free(tmp_oa); + RETURN(updates); +} /* FIXME: maybe we'll just make one node the authoritative attribute node, then * we can send this 'punch' to just the authoritative node and the nodes * that the punch will affect. */ -static int lov_punch(struct lustre_handle *conn, struct obdo *oa, +static int lov_punch(struct obd_export *exp, struct obdo *oa, struct lov_stripe_md *lsm, - obd_off start, obd_off end) + obd_off start, obd_off end, struct obd_trans_info *oti) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); + struct lov_request_set *set; struct lov_obd *lov; - struct lov_oinfo *loi; - int rc = 0, i; + struct list_head *pos; + struct lov_request *req; + int err = 0, rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + rc = lov_prep_punch_set(exp, oa, lsm, start, end, oti, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + req = list_entry(pos, struct lov_request, rq_link); + + rc = obd_punch(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, + NULL, req->rq_extent.start, + req->rq_extent.end, NULL); + err = lov_update_punch_set(set, req, rc); + if (err) { + CERROR("error: punch objid "LPX64" subobj "LPX64 + " on OST idx %d: rc = %d\n", set->set_oa->o_id, + req->rq_oa->o_id, req->rq_idx, rc); + if (!rc) + rc = err; + } } + err = lov_fini_punch_set(set); + if (!rc) + rc = err; + RETURN(rc); +} + +static int lov_sync(struct obd_export *exp, struct obdo *oa, + struct lov_stripe_md *lsm, obd_off start, obd_off end) +{ + struct lov_request_set *set; + struct lov_obd *lov; + struct list_head *pos; + struct lov_request *req; + int err = 0, rc = 0; + ENTRY; - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); - } - if (!export || !export->exp_obd) + if (!exp->exp_obd) RETURN(-ENODEV); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - obd_off starti = lov_stripe_offset(lsm, start, i); - obd_off endi = lov_stripe_offset(lsm, end, i); - int err; + lov = &exp->exp_obd->u.lov; + rc = lov_prep_sync_set(exp, oa, lsm, start, end, &set); + if (rc) + RETURN(rc); - if (starti == endi) - continue; - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; + list_for_each (pos, &set->set_list) { + req = list_entry(pos, struct lov_request, rq_link); - err = obd_punch(&lov->tgts[loi->loi_ost_idx].conn, &tmp, NULL, - starti, endi); + rc = obd_sync(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, + NULL, req->rq_extent.start, req->rq_extent.end); + err = lov_update_common_set(set, req, rc); if (err) { - CERROR("Error punch objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); + CERROR("error: fsync objid "LPX64" subobj "LPX64 + " on OST idx %d: rc = %d\n", set->set_oa->o_id, + req->rq_oa->o_id, req->rq_idx, rc); if (!rc) rc = err; } } + err = lov_fini_sync_set(set); + if (!rc) + rc = err; RETURN(rc); } -static int lov_osc_brw_callback(struct io_cb_data *cbd, int err, int phase) +static int lov_brw_check(struct lov_obd *lov, struct obdo *oa, + struct lov_stripe_md *lsm, + obd_count oa_bufs, struct brw_page *pga) { - int ret = 0; + int i, rc = 0; ENTRY; - if (phase == CB_PHASE_START) - RETURN(0); + /* The caller just wants to know if there's a chance that this + * I/O can succeed */ + for (i = 0; i < oa_bufs; i++) { + int stripe = lov_stripe_number(lsm, pga[i].disk_offset); + int ost = lsm->lsm_oinfo[stripe].loi_ost_idx; + obd_off start, end; - if (phase == CB_PHASE_FINISH) { - if (err) - cbd->err = err; - if (atomic_dec_and_test(&cbd->refcount)) - ret = cbd->cb(cbd->data, cbd->err, phase); - RETURN(ret); - } + if (!lov_stripe_intersects(lsm, i, pga[i].disk_offset, + pga[i].disk_offset + pga[i].count, + &start, &end)) + continue; - LBUG(); - return 0; + if (lov->tgts[ost].active == 0) { + CDEBUG(D_HA, "lov idx %d inactive\n", ost); + RETURN(-EIO); + } + rc = obd_brw(OBD_BRW_CHECK, lov->tgts[ost].ltd_exp, oa, + NULL, 1, &pga[i], NULL); + if (rc) + break; + } + RETURN(rc); } -static inline int lov_brw(int cmd, struct lustre_handle *conn, - struct lov_stripe_md *lsm, obd_count oa_bufs, - struct brw_page *pga, - brw_callback_t callback, struct io_cb_data *cbd) +static int lov_brw(int cmd, struct obd_export *exp, struct obdo *src_oa, + struct lov_stripe_md *lsm, obd_count oa_bufs, + struct brw_page *pga, struct obd_trans_info *oti) { - int stripe_count = lsm->lsm_stripe_count; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct { - int bufct; - int index; - int subcount; - struct lov_stripe_md lsm; - int ost_idx; - } *stripeinfo, *si, *si_last; - struct brw_page *ioarr; - int rc, i; - struct io_cb_data *our_cb; - struct lov_oinfo *loi; - int *where; + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov = &exp->exp_obd->u.lov; + int err, rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); + + if (cmd == OBD_BRW_CHECK) { + rc = lov_brw_check(lov, src_oa, lsm, oa_bufs, pga); + RETURN(rc); + } + + rc = lov_prep_brw_set(exp, src_oa, lsm, oa_bufs, pga, oti, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + struct obd_export *sub_exp; + struct brw_page *sub_pga; + req = list_entry(pos, struct lov_request, rq_link); + + sub_exp = lov->tgts[req->rq_idx].ltd_exp; + sub_pga = set->set_pga + req->rq_pgaidx; + rc = obd_brw(cmd, sub_exp, req->rq_oa, req->rq_md, + req->rq_oabufs, sub_pga, oti); + if (rc) + break; + lov_update_common_set(set, req, rc); + } + + err = lov_fini_brw_set(set); + if (!rc) + rc = err; + RETURN(rc); +} + +static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data, + int rc) +{ + struct lov_request_set *lovset = (struct lov_request_set *)data; + ENTRY; + + if (rc) { + lovset->set_completes = 0; + lov_fini_brw_set(lovset); + } else { + rc = lov_fini_brw_set(lovset); } + + RETURN(rc); +} + +static int lov_brw_async(int cmd, struct obd_export *exp, struct obdo *oa, + struct lov_stripe_md *lsm, obd_count oa_bufs, + struct brw_page *pga, struct ptlrpc_request_set *set, + struct obd_trans_info *oti) +{ + struct lov_request_set *lovset; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov = &exp->exp_obd->u.lov; + int rc = 0; + ENTRY; - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); + + if (cmd == OBD_BRW_CHECK) { + rc = lov_brw_check(lov, oa, lsm, oa_bufs, pga); + RETURN(rc); } - lov = &export->exp_obd->u.lov; + rc = lov_prep_brw_set(exp, oa, lsm, oa_bufs, pga, oti, &lovset); + if (rc) + RETURN(rc); - our_cb = ll_init_cb(); - if (!our_cb) - RETURN(-ENOMEM); + list_for_each (pos, &lovset->set_list) { + struct obd_export *sub_exp; + struct brw_page *sub_pga; + req = list_entry(pos, struct lov_request, rq_link); + + sub_exp = lov->tgts[req->rq_idx].ltd_exp; + sub_pga = lovset->set_pga + req->rq_pgaidx; + rc = obd_brw_async(cmd, sub_exp, req->rq_oa, req->rq_md, + req->rq_oabufs, sub_pga, set, oti); + if (rc) + GOTO(out, rc); + lov_update_common_set(lovset, req, rc); + } + LASSERT(rc == 0); + LASSERT(set->set_interpret == NULL); + set->set_interpret = (set_interpreter_func)lov_brw_interpret; + set->set_arg = (void *)lovset; + + RETURN(rc); +out: + lov_fini_brw_set(lovset); + RETURN(rc); +} - OBD_ALLOC(stripeinfo, stripe_count * sizeof(*stripeinfo)); - if (!stripeinfo) - GOTO(out_cbdata, rc = -ENOMEM); +static int lov_ap_make_ready(void *data, int cmd) +{ + struct lov_async_page *lap = LAP_FROM_COOKIE(data); - OBD_ALLOC(where, sizeof(*where) * oa_bufs); - if (!where) - GOTO(out_sinfo, rc = -ENOMEM); + return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd); +} +static int lov_ap_refresh_count(void *data, int cmd) +{ + struct lov_async_page *lap = LAP_FROM_COOKIE(data); - OBD_ALLOC(ioarr, sizeof(*ioarr) * oa_bufs); - if (!ioarr) - GOTO(out_where, rc = -ENOMEM); + return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data, + cmd); +} +static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa) +{ + struct lov_async_page *lap = LAP_FROM_COOKIE(data); - /* This is the only race-free way I can think of to get the refcount - * correct. -phil */ - atomic_set(&our_cb->refcount, 0); - our_cb->cb = callback; - our_cb->data = cbd; + lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa); + /* XXX woah, shouldn't we be altering more here? size? */ + oa->o_id = lap->lap_loi_id; +} - for (i = 0; i < oa_bufs; i++) { - where[i] = lov_stripe_number(lsm, pga[i].off); - if (stripeinfo[where[i]].bufct++ == 0) - atomic_inc(&our_cb->refcount); +static void lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc) +{ + struct lov_async_page *lap = LAP_FROM_COOKIE(data); + + /* in a raid1 regime this would down a count of many ios + * in flight, onl calling the caller_ops completion when all + * the raid1 ios are complete */ + lap->lap_caller_ops->ap_completion(lap->lap_caller_data, cmd, oa, rc); +} + +static struct obd_async_page_ops lov_async_page_ops = { + .ap_make_ready = lov_ap_make_ready, + .ap_refresh_count = lov_ap_refresh_count, + .ap_fill_obdo = lov_ap_fill_obdo, + .ap_completion = lov_ap_completion, +}; + +static int lov_prep_async_page(struct obd_export *exp, + struct lov_stripe_md *lsm, + struct lov_oinfo *loi, struct page *page, + obd_off offset, struct obd_async_page_ops *ops, + void *data, void **res) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_async_page *lap; + int rc, stripe; + ENTRY; + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + LASSERT(loi == NULL); + + stripe = lov_stripe_number(lsm, offset); + loi = &lsm->lsm_oinfo[stripe]; + + if (obd_uuid_empty(&lov->tgts[loi->loi_ost_idx].uuid)) + RETURN(-EIO); + if (lov->tgts[loi->loi_ost_idx].active == 0) + RETURN(-EIO); + if (lov->tgts[loi->loi_ost_idx].ltd_exp == NULL) { + CERROR("ltd_exp == NULL, but OST idx %d doesn't appear to be " + "deleted or inactive.\n", loi->loi_ost_idx); + RETURN(-EIO); } - for (i = 0, loi = lsm->lsm_oinfo, si_last = si = stripeinfo; - i < stripe_count; i++, loi++, si_last = si, si++) { - if (i > 0) - si->index = si_last->index + si_last->bufct; - si->lsm.lsm_object_id = loi->loi_id; - si->ost_idx = loi->loi_ost_idx; + OBD_ALLOC(lap, sizeof(*lap)); + if (lap == NULL) + RETURN(-ENOMEM); + + lap->lap_magic = LAP_MAGIC; + lap->lap_caller_ops = ops; + lap->lap_caller_data = data; + + /* FIXME handle multiple oscs after landing b_raid1 */ + lap->lap_stripe = stripe; + switch (lsm->lsm_pattern) { + case LOV_PATTERN_RAID0: + lov_stripe_offset(lsm, offset, lap->lap_stripe, + &lap->lap_sub_offset); + break; + case LOV_PATTERN_CMOBD: + lap->lap_sub_offset = offset; + break; + default: + LBUG(); } - for (i = 0; i < oa_bufs; i++) { - int which = where[i]; - int shift; + /* so the callback doesn't need the lsm */ + lap->lap_loi_id = loi->loi_id; - shift = stripeinfo[which].index + stripeinfo[which].subcount; - LASSERT(shift < oa_bufs); - ioarr[shift] = pga[i]; - ioarr[shift].off = lov_stripe_offset(lsm, pga[i].off, which); - stripeinfo[which].subcount++; + rc = obd_prep_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp, + lsm, loi, page, lap->lap_sub_offset, + &lov_async_page_ops, lap, + &lap->lap_sub_cookie); + if (rc) { + OBD_FREE(lap, sizeof(*lap)); + RETURN(rc); } + CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page, + lap->lap_sub_cookie, offset); + *res = lap; + RETURN(0); +} - for (i = 0, si = stripeinfo; i < stripe_count; i++, si++) { - int shift = si->index; +static int lov_queue_async_io(struct obd_export *exp, + struct lov_stripe_md *lsm, + struct lov_oinfo *loi, void *cookie, + int cmd, obd_off off, int count, + obd_flags brw_flags, obd_flags async_flags) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_async_page *lap; + int rc; - if (si->bufct) { - LASSERT(shift < oa_bufs); - /* XXX handle error returns here */ - obd_brw(cmd, &lov->tgts[si->ost_idx].conn, - &si->lsm, si->bufct, &ioarr[shift], - lov_osc_brw_callback, our_cb); + LASSERT(loi == NULL); + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + lap = LAP_FROM_COOKIE(cookie); + + loi = &lsm->lsm_oinfo[lap->lap_stripe]; + + rc = obd_queue_async_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm, + loi, lap->lap_sub_cookie, cmd, off, count, + brw_flags, async_flags); + RETURN(rc); +} + +static int lov_set_async_flags(struct obd_export *exp, + struct lov_stripe_md *lsm, + struct lov_oinfo *loi, void *cookie, + obd_flags async_flags) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_async_page *lap; + int rc; + + LASSERT(loi == NULL); + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + lap = LAP_FROM_COOKIE(cookie); + + loi = &lsm->lsm_oinfo[lap->lap_stripe]; + + rc = obd_set_async_flags(lov->tgts[loi->loi_ost_idx].ltd_exp, + lsm, loi, lap->lap_sub_cookie, async_flags); + RETURN(rc); +} + +static int lov_queue_group_io(struct obd_export *exp, + struct lov_stripe_md *lsm, + struct lov_oinfo *loi, + struct obd_io_group *oig, void *cookie, + int cmd, obd_off off, int count, + obd_flags brw_flags, obd_flags async_flags) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_async_page *lap; + int rc; + + LASSERT(loi == NULL); + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + lap = LAP_FROM_COOKIE(cookie); + + loi = &lsm->lsm_oinfo[lap->lap_stripe]; + + rc = obd_queue_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm, loi, + oig, lap->lap_sub_cookie, cmd, off, count, + brw_flags, async_flags); + RETURN(rc); +} + +/* this isn't exactly optimal. we may have queued sync io in oscs on + * all stripes, but we don't record that fact at queue time. so we + * trigger sync io on all stripes. */ +static int lov_trigger_group_io(struct obd_export *exp, + struct lov_stripe_md *lsm, + struct lov_oinfo *loi, + struct obd_io_group *oig) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + int rc = 0, i, err; + + LASSERT(loi == NULL); + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + loi = lsm->lsm_oinfo; + for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) { + if (lov->tgts[loi->loi_ost_idx].active == 0) { + CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx); + continue; } - } - rc = callback(cbd, 0, CB_PHASE_START); + err = obd_trigger_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp, + lsm, loi, oig); + if (rc == 0 && err != 0) + rc = err; + }; + RETURN(rc); +} + +static int lov_teardown_async_page(struct obd_export *exp, + struct lov_stripe_md *lsm, + struct lov_oinfo *loi, void *cookie) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_async_page *lap; + int rc; + + LASSERT(loi == NULL); + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); - OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs); - out_where: - OBD_FREE(where, sizeof(*where) * oa_bufs); - out_sinfo: - OBD_FREE(stripeinfo, stripe_count * sizeof(*stripeinfo)); - out_cbdata: - OBD_FREE(our_cb, sizeof(*our_cb)); + lap = LAP_FROM_COOKIE(cookie); + + loi = &lsm->lsm_oinfo[lap->lap_stripe]; + + rc = obd_teardown_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp, + lsm, loi, lap->lap_sub_cookie); + if (rc) { + CERROR("unable to teardown sub cookie %p: %d\n", + lap->lap_sub_cookie, rc); + RETURN(rc); + } + OBD_FREE(lap, sizeof(*lap)); RETURN(rc); } -static int lov_enqueue(struct lustre_handle *conn, struct lov_stripe_md *lsm, - struct lustre_handle *parent_lock, - __u32 type, void *cookie, int cookielen, __u32 mode, - int *flags, void *cb, void *data, int datalen, - struct lustre_handle *lockhs) +static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm, + __u32 type, ldlm_policy_data_t *policy, __u32 mode, + int *flags, void *bl_cb, void *cp_cb, void *gl_cb, + void *data,__u32 lvb_len, void *lvb_swabber, + struct lustre_handle *lockh) { - struct obd_export *export = class_conn2export(conn); + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lustre_handle *lov_lockhp; struct lov_obd *lov; - struct lov_oinfo *loi; - int rc = 0, i; + ldlm_error_t rc; + int save_flags = *flags; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); + + /* we should never be asked to replay a lock this way. */ + LASSERT((*flags & LDLM_FL_REPLAY) == 0); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + rc = lov_prep_enqueue_set(exp, lsm, policy, mode, lockh, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + ldlm_policy_data_t sub_policy; + req = list_entry(pos, struct lov_request, rq_link); + lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; + LASSERT(lov_lockhp); + + *flags = save_flags; + sub_policy.l_extent.start = req->rq_extent.start; + sub_policy.l_extent.end = req->rq_extent.end; + + rc = obd_enqueue(lov->tgts[req->rq_idx].ltd_exp, req->rq_md, + type, &sub_policy, mode, flags, bl_cb, + cp_cb, gl_cb, data, lvb_len, lvb_swabber, + lov_lockhp); + rc = lov_update_enqueue_set(set, req, rc, save_flags); + if (rc != ELDLM_OK) + break; } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); + lov_fini_enqueue_set(set, mode); + RETURN(rc); +} + +static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm, + __u32 type, ldlm_policy_data_t *policy, __u32 mode, + int *flags, void *data, struct lustre_handle *lockh) +{ + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lustre_handle *lov_lockhp; + int lov_flags, rc = 0; + ENTRY; + + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); - } - if (!export || !export->exp_obd) + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - struct ldlm_extent *extent = (struct ldlm_extent *)cookie; - struct ldlm_extent sub_ext; - struct lov_stripe_md submd; - - sub_ext.start = lov_stripe_offset(lsm, extent->start, i); - sub_ext.end = lov_stripe_offset(lsm, extent->end, i); - if (sub_ext.start == sub_ext.end) - continue; + lov = &exp->exp_obd->u.lov; + rc = lov_prep_match_set(exp, lsm, policy, mode, lockh, &set); + if (rc) + RETURN(rc); - submd.lsm_object_id = loi->loi_id; - /* XXX submd lsm_mds_easize should be that from the subobj, - * and the subobj should get it opaquely from the LOV. - */ - submd.lsm_mds_easize = lov_mds_md_size(lsm->lsm_ost_count); - submd.lsm_stripe_count = 0; - /* XXX submd is not fully initialized here */ - rc = obd_enqueue(&(lov->tgts[loi->loi_ost_idx].conn), &submd, - parent_lock, type, &sub_ext, sizeof(sub_ext), - mode, flags, cb, data, datalen, &(lockhs[i])); - // XXX add a lock debug statement here - if (rc) - CERROR("Error enqueue objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", lsm->lsm_object_id, - loi->loi_id, loi->loi_ost_idx, rc); + list_for_each (pos, &set->set_list) { + ldlm_policy_data_t sub_policy; + req = list_entry(pos, struct lov_request, rq_link); + lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; + LASSERT(lov_lockhp); + + sub_policy.l_extent.start = req->rq_extent.start; + sub_policy.l_extent.end = req->rq_extent.end; + lov_flags = *flags; + + rc = obd_match(lov->tgts[req->rq_idx].ltd_exp, req->rq_md, + type, &sub_policy, mode, &lov_flags, data, + lov_lockhp); + rc = lov_update_match_set(set, req, rc); + if (rc != 1) + break; } + lov_fini_match_set(set, mode, *flags); RETURN(rc); } -static int lov_cancel(struct lustre_handle *conn, struct lov_stripe_md *lsm, - __u32 mode, struct lustre_handle *lockhs) +static int lov_change_cbdata(struct obd_export *exp, + struct lov_stripe_md *lsm, ldlm_iterator_t it, + void *data) { - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; struct lov_oinfo *loi; int rc = 0, i; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + if (lsm_bad_magic(lsm)) RETURN(-EINVAL); - } - - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); - } - if (!export || !export->exp_obd) + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - lov = &export->exp_obd->u.lov; + LASSERT(lsm->lsm_object_gr > 0); + + lov = &exp->exp_obd->u.lov; for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { struct lov_stripe_md submd; - - if (lockhs[i].addr == 0) + if (lov->tgts[loi->loi_ost_idx].active == 0) { + CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx); continue; + } submd.lsm_object_id = loi->loi_id; - submd.lsm_mds_easize = lov_mds_md_size(lsm->lsm_ost_count); + submd.lsm_object_gr = lsm->lsm_object_gr; submd.lsm_stripe_count = 0; - rc = obd_cancel(&lov->tgts[loi->loi_ost_idx].conn, &submd, - mode, &lockhs[i]); - if (rc) - CERROR("Error cancel objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", lsm->lsm_object_id, - loi->loi_id, loi->loi_ost_idx, rc); + rc = obd_change_cbdata(lov->tgts[loi->loi_ost_idx].ltd_exp, + &submd, it, data); } RETURN(rc); } -static int lov_cancel_unused(struct lustre_handle *conn, - struct lov_stripe_md *lsm, int local_only) +static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm, + __u32 mode, struct lustre_handle *lockh) +{ + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lustre_handle *lov_lockhp; + int err = 0, rc = 0; + ENTRY; + + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + LASSERT(lsm->lsm_object_gr > 0); + + LASSERT(lockh); + lov = &exp->exp_obd->u.lov; + rc = lov_prep_cancel_set(exp, lsm, mode, lockh, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + req = list_entry(pos, struct lov_request, rq_link); + lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; + + rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md, + mode, lov_lockhp); + rc = lov_update_common_set(set, req, rc); + if (rc) { + CERROR("error: cancel objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + lsm->lsm_object_id, + req->rq_md->lsm_object_id, req->rq_idx, rc); + err = rc; + } + + } + lov_fini_cancel_set(set); + RETURN(err); +} + +static int lov_cancel_unused(struct obd_export *exp, + struct lov_stripe_md *lsm, + int flags, void *opaque) { - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; struct lov_oinfo *loi; int rc = 0, i; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea for lock cancellation\n"); - RETURN(-EINVAL); + lov = &exp->exp_obd->u.lov; + if (lsm == NULL) { + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + int err = obd_cancel_unused(lov->tgts[i].ltd_exp, + NULL, flags, opaque); + if (!rc) + rc = err; + } + RETURN(rc); } - if (!export || !export->exp_obd) + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); + + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - lov = &export->exp_obd->u.lov; + LASSERT(lsm->lsm_object_gr > 0); + for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { struct lov_stripe_md submd; + int err; + + if (lov->tgts[loi->loi_ost_idx].active == 0) + CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx); submd.lsm_object_id = loi->loi_id; - submd.lsm_mds_easize = lov_mds_md_size(lsm->lsm_ost_count); + submd.lsm_object_gr = lsm->lsm_object_gr; submd.lsm_stripe_count = 0; - rc = obd_cancel_unused(&lov->tgts[loi->loi_ost_idx].conn, - &submd, local_only); - if (rc) - CERROR("Error cancel unused objid "LPX64" subobj "LPX64 + err = obd_cancel_unused(lov->tgts[loi->loi_ost_idx].ltd_exp, + &submd, flags, opaque); + if (err && lov->tgts[loi->loi_ost_idx].active) { + CERROR("error: cancel unused objid "LPX64" subobj "LPX64 " on OST idx %d: rc = %d\n", lsm->lsm_object_id, - loi->loi_id, loi->loi_ost_idx, rc); + loi->loi_id, loi->loi_ost_idx, err); + if (!rc) + rc = err; + } } RETURN(rc); } -static int lov_statfs(struct lustre_handle *conn, struct obd_statfs *osfs) +#define LOV_U64_MAX ((__u64)~0ULL) +#define LOV_SUM_MAX(tot, add) \ + do { \ + if ((tot) + (add) < (tot)) \ + (tot) = LOV_U64_MAX; \ + else \ + (tot) += (add); \ + } while(0) + +static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs, + unsigned long max_age) { - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; + struct lov_obd *lov = &obd->u.lov; struct obd_statfs lov_sfs; int set = 0; int rc = 0; int i; ENTRY; - if (!export || !export->exp_obd) - RETURN(-ENODEV); - - lov = &export->exp_obd->u.lov; /* We only get block data from the OBD */ for (i = 0; i < lov->desc.ld_tgt_count; i++) { int err; + if (!lov->tgts[i].active) { + CDEBUG(D_HA, "lov idx %d inactive\n", i); + continue; + } - err = obd_statfs(&lov->tgts[i].conn, &lov_sfs); + err = obd_statfs(class_exp2obd(lov->tgts[i].ltd_exp), &lov_sfs, + max_age); if (err) { - CERROR("Error statfs OSC %s idx %d: err = %d\n", - lov->tgts[i].uuid, i, err); - if (!rc) + if (lov->tgts[i].active && !rc) rc = err; - continue; /* XXX or break? - probably OK to continue */ + continue; } + if (!set) { memcpy(osfs, &lov_sfs, sizeof(lov_sfs)); set = 1; @@ -1122,78 +1829,491 @@ static int lov_statfs(struct lustre_handle *conn, struct obd_statfs *osfs) * if one of the OBDs has no more objects left) * - could be sum if we stripe whole objects * - could be average, just to give a nice number - * - we just pick first OST and hope it is enough - sfs->f_ffree += lov_sfs.f_ffree; + * + * To give a "reasonable" (if not wholly accurate) + * number, we divide the total number of free objects + * by expected stripe count (watch out for overflow). */ + LOV_SUM_MAX(osfs->os_files, lov_sfs.os_files); + LOV_SUM_MAX(osfs->os_ffree, lov_sfs.os_ffree); } } + + if (set) { + __u32 expected_stripes = lov->desc.ld_default_stripe_count ? + lov->desc.ld_default_stripe_count : + lov->desc.ld_active_tgt_count; + + if (osfs->os_files != LOV_U64_MAX) + do_div(osfs->os_files, expected_stripes); + if (osfs->os_ffree != LOV_U64_MAX) + do_div(osfs->os_ffree, expected_stripes); + } else if (!rc) + rc = -EIO; + RETURN(rc); } -static int lov_iocontrol(long cmd, struct lustre_handle *conn, int len, +static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg) { - struct obd_device *obddev = class_conn2obd(conn); - struct obd_ioctl_data *data = karg; - int rc; + struct obd_device *obddev = class_exp2obd(exp); + struct lov_obd *lov = &obddev->u.lov; + int i, rc, count = lov->desc.ld_tgt_count; + struct obd_uuid *uuidp; ENTRY; - if (_IOC_TYPE(cmd) != IOC_LOV_TYPE || - _IOC_NR(cmd) < IOC_LOV_MIN_NR || _IOC_NR(cmd) > IOC_LOV_MAX_NR) { - CDEBUG(D_IOCTL, "invalid ioctl (type %ld, nr %ld, size %ld)\n", - _IOC_TYPE(cmd), _IOC_NR(cmd), _IOC_SIZE(cmd)); - RETURN(-ENOTTY); - } - switch (cmd) { - case IOC_LOV_SET_OSC_ACTIVE: - rc = lov_set_osc_active(&obddev->u.lov, data->ioc_inlbuf1, - data->ioc_offset); + case OBD_IOC_LOV_GET_CONFIG: { + struct obd_ioctl_data *data = karg; + struct lov_tgt_desc *tgtdesc; + struct lov_desc *desc; + char *buf = NULL; + __u32 *genp; + + buf = NULL; + len = 0; + if (obd_ioctl_getdata(&buf, &len, (void *)uarg)) + RETURN(-EINVAL); + + data = (struct obd_ioctl_data *)buf; + + if (sizeof(*desc) > data->ioc_inllen1) { + obd_ioctl_freedata(buf, len); + RETURN(-EINVAL); + } + + if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) { + obd_ioctl_freedata(buf, len); + RETURN(-EINVAL); + } + + if (sizeof(__u32) * count > data->ioc_inllen3) { + obd_ioctl_freedata(buf, len); + RETURN(-EINVAL); + } + + desc = (struct lov_desc *)data->ioc_inlbuf1; + memcpy(desc, &(lov->desc), sizeof(*desc)); + + uuidp = (struct obd_uuid *)data->ioc_inlbuf2; + genp = (__u32 *)data->ioc_inlbuf3; + tgtdesc = lov->tgts; + /* the uuid will be empty for deleted OSTs */ + for (i = 0; i < count; i++, uuidp++, genp++, tgtdesc++) { + obd_str2uuid(uuidp, (char *)tgtdesc->uuid.uuid); + *genp = tgtdesc->ltd_gen; + } + + rc = copy_to_user((void *)uarg, buf, len); + if (rc) + rc = -EFAULT; + obd_ioctl_freedata(buf, len); break; - default: - RETURN(-ENOTTY); + } + case LL_IOC_LOV_SETSTRIPE: + rc = lov_setstripe(exp, karg, uarg); + break; + case LL_IOC_LOV_GETSTRIPE: + rc = lov_getstripe(exp, karg, uarg); + break; + case LL_IOC_LOV_SETEA: + rc = lov_setea(exp, karg, uarg); + break; + default: { + int set = 0; + if (count == 0) + RETURN(-ENOTTY); + rc = 0; + for (i = 0; i < count; i++) { + int err; + + /* OST was deleted */ + if (obd_uuid_empty(&lov->tgts[i].uuid)) + continue; + + err = obd_iocontrol(cmd, lov->tgts[i].ltd_exp, + len, karg, uarg); + if (err) { + if (lov->tgts[i].active) { + CERROR("error: iocontrol OSC %s on OST " + "idx %d cmd %x: err = %d\n", + lov->tgts[i].uuid.uuid, i, + cmd, err); + if (!rc) + rc = err; + } + } else + set = 1; + } + if (!set && !rc) + rc = -EIO; + } } RETURN(rc); } -struct obd_ops lov_obd_ops = { - o_setup: lov_setup, - o_connect: lov_connect, - o_disconnect: lov_disconnect, - o_create: lov_create, - o_destroy: lov_destroy, - o_getattr: lov_getattr, - o_setattr: lov_setattr, - o_statfs: lov_statfs, - o_open: lov_open, - o_close: lov_close, - o_brw: lov_brw, - o_punch: lov_punch, - o_enqueue: lov_enqueue, - o_cancel: lov_cancel, - o_cancel_unused: lov_cancel_unused, - o_iocontrol: lov_iocontrol +static int lov_get_info(struct obd_export *exp, __u32 keylen, + void *key, __u32 *vallen, void *val) +{ + struct obd_device *obddev = class_exp2obd(exp); + struct lov_obd *lov = &obddev->u.lov; + int i; + ENTRY; + + if (!vallen || !val) + RETURN(-EFAULT); + + if (keylen > strlen("lock_to_stripe") && + strcmp(key, "lock_to_stripe") == 0) { + struct { + char name[16]; + struct ldlm_lock *lock; + struct lov_stripe_md *lsm; + } *data = key; + struct lov_oinfo *loi; + struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name; + __u32 *stripe = val; + + if (*vallen < sizeof(*stripe)) + RETURN(-EFAULT); + *vallen = sizeof(*stripe); + + /* XXX This is another one of those bits that will need to + * change if we ever actually support nested LOVs. It uses + * the lock's export to find out which stripe it is. */ + /* XXX - it's assumed all the locks for deleted OSTs have + * been cancelled. Also, the export for deleted OSTs will + * be NULL and won't match the lock's export. */ + for (i = 0, loi = data->lsm->lsm_oinfo; + i < data->lsm->lsm_stripe_count; + i++, loi++) { + if (lov->tgts[loi->loi_ost_idx].ltd_exp == + data->lock->l_conn_export && + loi->loi_id == res_id->name[0] && + loi->loi_gr == res_id->name[2]) { + *stripe = i; + RETURN(0); + } + } + LDLM_ERROR(data->lock, "lock on inode without such object\n"); + dump_lsm(D_ERROR, data->lsm); + RETURN(-ENXIO); + } else if (keylen >= strlen("size_to_stripe") && + strcmp(key, "size_to_stripe") == 0) { + struct { + int stripe_number; + __u64 size; + struct lov_stripe_md *lsm; + } *data = val; + + if (*vallen < sizeof(*data)) + RETURN(-EFAULT); + + data->size = lov_size_to_stripe(data->lsm, data->size, + data->stripe_number); + RETURN(0); + } else if (keylen >= strlen("last_id") && strcmp(key, "last_id") == 0) { + __u32 size = sizeof(obd_id); + obd_id *ids = val; + int rc = 0; + + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + if (!lov->tgts[i].active) + continue; + rc = obd_get_info(lov->tgts[i].ltd_exp, + keylen, key, &size, &(ids[i])); + if (rc != 0) + RETURN(rc); + } + RETURN(0); + } else if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) { + struct lov_desc *desc_ret = val; + *desc_ret = lov->desc; + + RETURN(0); + } + + RETURN(-EINVAL); +} + +static int lov_set_info(struct obd_export *exp, obd_count keylen, + void *key, obd_count vallen, void *val) +{ + struct obd_device *obddev = class_exp2obd(exp); + struct lov_obd *lov = &obddev->u.lov; + int i, rc = 0, err; + ENTRY; + +#define KEY_IS(str) \ + (keylen == strlen(str) && memcmp(key, str, keylen) == 0) + + if (KEY_IS("async")) { + struct lov_desc *desc = &lov->desc; + struct lov_tgt_desc *tgts = lov->tgts; + + if (vallen != sizeof(int)) + RETURN(-EINVAL); + lov->async = *((int*) val); + + for (i = 0; i < desc->ld_tgt_count; i++, tgts++) { + struct obd_uuid *tgt_uuid = &tgts->uuid; + struct obd_device *tgt_obd; + + tgt_obd = class_find_client_obd(tgt_uuid, + LUSTRE_OSC_NAME, + &obddev->obd_uuid); + if (!tgt_obd) { + CERROR("Target %s not attached\n", + tgt_uuid->uuid); + if (!rc) + rc = -EINVAL; + continue; + } + + err = obd_set_info(tgt_obd->obd_self_export, + keylen, key, vallen, val); + if (err) { + CERROR("Failed to set async on target %s\n", + tgt_obd->obd_name); + if (!rc) + rc = err; + } + } + RETURN(rc); + } + + if (KEY_IS("mds_conn")) { + if (vallen != sizeof(__u32)) + RETURN(-EINVAL); + } else if (KEY_IS("unlinked") || KEY_IS("unrecovery")) { + if (vallen != 0) + RETURN(-EINVAL); + } else if (KEY_IS("sec")) { + struct lov_tgt_desc *tgt; + struct obd_export *exp; + int rc = 0, err, i; + + spin_lock(&lov->lov_lock); + for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; + i++, tgt++) { + exp = tgt->ltd_exp; + /* during setup time the connections to osc might + * haven't been established. + */ + if (exp == NULL) { + struct obd_device *tgt_obd; + + tgt_obd = class_find_client_obd(&tgt->uuid, + LUSTRE_OSC_NAME, + &obddev->obd_uuid); + if (!tgt_obd) { + CERROR("can't set security flavor, " + "device %s not attached?\n", + tgt->uuid.uuid); + rc = -EINVAL; + continue; + } + exp = tgt_obd->obd_self_export; + } + + err = obd_set_info(exp, keylen, key, vallen, val); + if (!rc) + rc = err; + } + spin_unlock(&lov->lov_lock); + + RETURN(rc); + } else if (KEY_IS("flush_cred")) { + struct lov_tgt_desc *tgt; + int rc = 0, i; + + for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; + i++, tgt++) { + if (!tgt->ltd_exp) + continue; + rc = obd_set_info(tgt->ltd_exp, + keylen, key, vallen, val); + if (rc) + RETURN(rc); + } + + RETURN(0); + } else { + RETURN(-EINVAL); + } + + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + if (val && !obd_uuid_equals(val, &lov->tgts[i].uuid)) + continue; + + if (!val && !lov->tgts[i].active) + continue; + + err = obd_set_info(lov->tgts[i].ltd_exp, + keylen, key, vallen, val); + if (!rc) + rc = err; + } + RETURN(rc); +#undef KEY_IS +} + +#if 0 +struct lov_multi_wait { + struct ldlm_lock *lock; + wait_queue_t wait; + int completed; + int generation; }; +int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm, + struct lustre_handle *lockh) +{ + struct lov_lock_handles *lov_lockh = NULL; + struct lustre_handle *lov_lockhp; + struct lov_obd *lov; + struct lov_oinfo *loi; + struct lov_multi_wait *queues; + int rc = 0, i; + ENTRY; -#define LOV_VERSION "v0.1" + if (lsm_bad_magic(lsm)) + RETURN(-EINVAL); -static int __init lov_init(void) + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + LASSERT(lockh != NULL); + if (lsm->lsm_stripe_count > 1) { + lov_lockh = lov_handle2llh(lockh); + if (lov_lockh == NULL) { + CERROR("LOV: invalid lov lock handle %p\n", lockh); + RETURN(-EINVAL); + } + + lov_lockhp = lov_lockh->llh_handles; + } else { + lov_lockhp = lockh; + } + + OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues)); + if (queues == NULL) + GOTO(out, rc = -ENOMEM); + + lov = &exp->exp_obd->u.lov; + for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; + i++, loi++, lov_lockhp++) { + struct ldlm_lock *lock; + struct obd_device *obd; + unsigned long irqflags; + + lock = ldlm_handle2lock(lov_lockhp); + if (lock == NULL) { + CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n", + loi->loi_ost_idx, loi->loi_id); + queues[i].completed = 1; + continue; + } + + queues[i].lock = lock; + init_waitqueue_entry(&(queues[i].wait), current); + add_wait_queue(lock->l_waitq, &(queues[i].wait)); + + obd = class_exp2obd(lock->l_conn_export); + if (obd != NULL) + imp = obd->u.cli.cl_import; + if (imp != NULL) { + spin_lock_irqsave(&imp->imp_lock, irqflags); + queues[i].generation = imp->imp_generation; + spin_unlock_irqrestore(&imp->imp_lock, irqflags); + } + } + + lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait, + interrupted_completion_wait, &lwd); + rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi); + + for (i = 0; i < lsm->lsm_stripe_count; i++) + remove_wait_queue(lock->l_waitq, &(queues[i].wait)); + + if (rc == -EINTR || rc == -ETIMEDOUT) { + + + } + + out: + if (lov_lockh != NULL) + lov_llh_put(lov_lockh); + RETURN(rc); +} +#endif + +struct obd_ops lov_obd_ops = { + .o_owner = THIS_MODULE, + .o_attach = lov_attach, + .o_detach = lov_detach, + .o_setup = lov_setup, + .o_cleanup = lov_cleanup, + .o_process_config = lov_process_config, + .o_connect = lov_connect, + .o_disconnect = lov_disconnect, + .o_statfs = lov_statfs, + .o_packmd = lov_packmd, + .o_unpackmd = lov_unpackmd, + .o_revalidate_md = lov_revalidate_md, + .o_create = lov_create, + .o_destroy = lov_destroy, + .o_getattr = lov_getattr, + .o_getattr_async = lov_getattr_async, + .o_setattr = lov_setattr, + .o_brw = lov_brw, + .o_brw_async = lov_brw_async, + .o_prep_async_page = lov_prep_async_page, + .o_queue_async_io = lov_queue_async_io, + .o_set_async_flags = lov_set_async_flags, + .o_queue_group_io = lov_queue_group_io, + .o_trigger_group_io = lov_trigger_group_io, + .o_teardown_async_page = lov_teardown_async_page, + .o_adjust_kms = lov_adjust_kms, + .o_punch = lov_punch, + .o_sync = lov_sync, + .o_enqueue = lov_enqueue, + .o_match = lov_match, + .o_change_cbdata = lov_change_cbdata, + .o_cancel = lov_cancel, + .o_cancel_unused = lov_cancel_unused, + .o_iocontrol = lov_iocontrol, + .o_get_info = lov_get_info, + .o_set_info = lov_set_info, + .o_llog_init = lov_llog_init, + .o_llog_finish = lov_llog_finish, + .o_notify = lov_notify, +}; + +int __init lov_init(void) { - printk(KERN_INFO "Lustre Logical Object Volume driver " LOV_VERSION - ", info@clusterfs.com\n"); - return class_register_type(&lov_obd_ops, OBD_LOV_DEVICENAME); + struct lprocfs_static_vars lvars; + int rc; + ENTRY; + + lprocfs_init_vars(lov, &lvars); + rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars, + OBD_LOV_DEVICENAME); + RETURN(rc); } -static void __exit lov_exit(void) +#ifdef __KERNEL__ +static void /*__exit*/ lov_exit(void) { class_unregister_type(OBD_LOV_DEVICENAME); } MODULE_AUTHOR("Cluster File Systems, Inc. "); -MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver v0.1"); +MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver"); MODULE_LICENSE("GPL"); module_init(lov_init); module_exit(lov_exit); +#endif